Apparently this is a rather known issue: Vagrant/VirtualBox/Apache2 Strange Cache Behaviour, http://smotko.si/nginx-static-file-problem/, https://twitter.com/meinharrd/status/580098162716774400, that attempting to use VirtualBox with Nginx or Apache exhibits buggy behavior (modifying a file will update the contents, but not the length, so making it shorter leads to garbage at the end, and making it longer just truncates at the original length). The solution is always to disable sendfile.
So I disabled sendfile and this actually fixed my problem, but now I see all of those static files taking 2-3 seconds to load every time.
Has anyone seen this sort of behavior before? I'm specifically using Boot2Docker to run Nginx and an app in PHP
There is a known issue on github about volume performance in boot2docker: https://github.com/boot2docker/boot2docker/issues/593
In that issue there is an interesting link about a A productive development environment with Docker on OS X.
What about a shared folder in VirtualBox?
Don’t use this feature for nginx inside a VM, create a real shared resource on the host instead.
Related
I am having an issue with a relatively small openstack cluster deployed with kolla-ansible. The issue is that after a few days the controllers stop working. When I go into the docker container logs, I see in all of them that there are Too Many Open Files. I have tried changing limits.conf sysctl max files for processes and user. After all of that, the issue still shows up.
One interesting thing is that this was not happening until I had to reboot all of the controllers. I rebooted them because I needed to increase the amount of ram that they have after they died swapping. My first thought was that kolla-ansible is setting a configuration after running deploy, but I can't seem to find any point in the repo when kolla-ansible is changing ulimits or other.
Any theories what could cause this? Would it be related to increasing ram? Should I run reconfigure/deploy on each controller? I've tried looking in kolla-ansible's docs and forums and couldn't see where anyone else was having this issue.
Update this hasn't been fixed yet:
I submitted a bug report, https://bugs.launchpad.net/kolla-ansible/+bug/1901898
I don't know your used versions of Kolla-Ansible and your Linux, but your problem seems really related to this one:
On Ubuntu 16.04, please uninstall lxd and lxc packages. (An issue exists with cgroup mounts, mounts exponentially increasing when restarting container) (source: docs.openstack.org/kolla-ansible/4.0.0/quickstart.html)
I had this problem with the exponentially growing number of mount-pointers after the restart of my docker-containers too. My single-node test-deployment had become very slow based on this problem, but I can't remember at the moment, that I would had the same error with too many open files.
You can delete the packages with apt-get remove lxc-common lxcfs lxd lxd-client. I had done this fix together with a complete reinstallation of the kolla-ansible installation, so I don't know, if this also helps with an already existing installation. You should also use docker-ce instead of the docker from the apt-repos.
This was fixed with a workaround in bug https://bugs.launchpad.net/keystonemiddleware/+bug/1883659 problem was neutron server was keeping memcached connections open and not closing them until the memcached container reached too many files open. There is a work around mentioned in the bug link.
We have one particular site that is Symfony and uses the e-commerce bundle Sylius.
Our developers are trying to use Vagrant so we can have similar dev environments. We use Puphpet to generate the Vagrant instance and share the config file.
If we are working on the site/repo natively or on a staging server, all runs fine. Pages load in around 2-3 seconds.
When we are using Vagrant / Virtualbox, it's 30-35 seconds per page load.
So far we've tried
Allocating up to 6GB to the box
Giving up to 4 processors to the box
Turning on NFS for file sync
Turning off all other programs on computers running Vagrant / Virtualbox (chat, other browsers, etc)
None of those things made an impact on page load time.
I can provide 2 things. One is the load trace from Symfony: https://nimbus.everhelper.me/client/notes/share/708707/mvw707mckzm2wq4rlkzc
Since there is so much code to the puphpet config, I put it in a pastebin here: http://pastebin.com/7ciVA5FL
What is OS on a host machine?
My guess would be that file system is slow. Try to run an app outside of shared folder on the guest machine. If it will be fast, then you'll spot a problem at least.
NFS on *nix or mac should be fast enough, are you sure you've succeed to turn it on?
I had this pain once, and finally started to use unison instead of native vagrant's file sharing system (https://www.cis.upenn.edu/~bcpierce/unison/)
Have your tried:
http://www.whitewashing.de/2013/08/19/speedup_symfony2_on_vagrant_boxes.html
or http://jeremybarthe.com/2015/02/02/speed-up-vagrant-environment-symfony2/
I think the first one is already included in Sylius, but not sure.
Also, dynamic image resize/crop may be reading/writing in the host file system and maybe there's a way to also change that (using symlinks or similar)?
vagrant-winnfsd works fine for me for getting NFS to work on Windows.
There's a bug in Virtualbox that results in small, static files not being transmitted over shared-folders properly. This purported solution is to add "sendfile off;" to the server/location block in Nginx (or a corresponding fix to Apache/etc...) and reloading. This prevents gibberish from being received in your Javascript, CSS, etc..
However, it's not working for me. I'm definitely hitting the right server/location blocks, but, even after reconstructing the Vagrant instance (with the Virtualbox provider), I still get gibberish in my files.
Does anyone have any guesses at what could be going on?
Yes but this can not be solved, you should create a real share on the Host and net use to that share, VB shared folders are a cheap way to get stuff in a VM but don't use it for anything else.
My Meteor file change watcher is taking forever to detect my file changes and refresh the browser, sometimes even longer than a minute. This makes developing a real pain.
My Meteor is running inside an Ubuntu-VM. The projects folder lies in my OSX and is mounted inside the VM. So I'm aware that inotify/kqueue won't work, so Meteor should fallback to stat polling.
I even set the environment variables according to this post, but the behavior is still the same.
METEOR_WATCH_FORCE_POLLING=true
METEOR_WATCH_POLLING_INTERVAL_MS=500
Is there any way to fix this annoying behavior?
The folder from OSX is mounted as a nfs share btw.
Update:
I did some testing and there is no difference if the application has a big amount of packages or is taking long to build, even with the very basic app you get after meteor create I still get the same behavior.
If I change a file in the VM (so that inotify works) the refresh is happening instantly.
I've have apps in production that incrementally become slower when adding packages, both 3rd party and private packages. I also discovered that adding 3rd libs directly on the client/lib increases the reloading time.
I'm not sure if Meteor 1.0.2 actually solved the problem of watching the directory efficiently.
What version of Meteor are you using?
Much of this lag-reloading issue was addressed on Meteor 1.0.2. While it still takes some time, I would say was ~5x faster on my experience.
I've run into an issue with Saltstack version 2014.7.0, where I cannot get network information from Salt.
If I run:
salt-call network.ip_addrs
I get:
Function network.ip_addrs is not available
This only seems to happen on some of my hosts. It seems to effect the almost all of the functions in salt.modules.network, but everything else works as expected.
I suspect there's something in my environment to blame. I am running salt within a CentOS 7 docker container. I followed these instructions to get Systemd running under Docker, and it seems to be functioning just fine, so I don't think that's the issue, but I wouldn't be surprised if it's related. I'm using Docker as a development environment, but I will be using these formula to orchestrate virtual machines in production.
Has anyone encountered the network module not being loaded properly? Is there something that needs to be available for that module to be accessible?
I have other mechanisms to get the IP address, but none that are as easy to work with in other salt formulas.
It turns out my problem was that I had my own custom module called "network" which was obscuring the upstream network module.
I'm pretty sure this was working at some point in the past, so I'm wondering if there might have been a change to salt in a more recent version that would cause it to conflict at a module level instead of merging methods from different modules of the same name, but I suppose it's possible that it never worked.